CRISPRi library V2, data processing pipeline

---

1 Description

This R notebook details the data processing and visualization for growth competition experiments with a CRISPRi sgRNA library. The library contains around 20,000 unique sgRNA repression mutants tailored for the cyanobacterium Synechocystis sp. PCC6803. This library is the second version (therefore “V2”) of an sgRNA library for Synechocystis, containing five instead of only two sgRNAs per gene. In some cases, genes or ncRNAs are so short that it is not possible to design a maximum of five individual sgRNAs.

The first iteration of the Synechocystis sgRNA library was published in Nature Communications, 2020.

2 Prerequisites

Load required packages.

suppressPackageStartupMessages({
  library(tidyverse)
  library(ggrepel)
  library(lattice)
  library(latticeExtra)
  library(latticetools)
  library(scales)
  library(dendextend)
  library(vegan)
  library(tsne)
  library(KEGGREST)
  library(limma)
  library(corrplot)
  library(kableExtra)
  library(grid)
  library(ggpubr)
})

Define global figure style, default colors, and a plot saving function.

3 Quality control

3.1 Data import

Load raw data. The main table contains already normalized quantification of all sgRNAs, fold change, multiple hypothesis corrected p-values, and fitness score. Contrary to the processing of our first CRISPRi library V1, much of the functionality from the notebook was transferred into the new CRISPRi library pipeline on github.

# load first seq run
load("../data/input/DESeq2_result.Rdata")
df_main <- DESeq_result_table

# load second seq run
load("../data/input/DESeq2_result_2.Rdata")
df_main <- bind_rows(df_main, DESeq_result_table)

# remove single results table
rm(DESeq_result_table)

3.2 Data annotation

Different annotation columns are added to the main data frame, including a short sgRNA identifier (excluding the position on the gene), an sgRNA index (1 to 5), and genome annotation from Uniprot. The Uniprot data is dynamically downloaded for every update of this pipeline using their very simple API (read_tsv("https://www.uniprot.org/uniprot/?query=taxonomy:1111708&format=tab")). The full list of columns that can be queried is available here. Pathway annotation from KEGG is later in the pipeline added using the KEGGREST package.

df_main <- df_main %>%
  # correct an error in sgRNA naming
  mutate(sgRNA = gsub('”', '2', sgRNA)) %>%
  # split sgRNA names into target gene and position
  separate(sgRNA, into = c("sgRNA_target", "sgRNA_position"), sep = "\\|",
    remove = FALSE) %>%
  
  # add sgRNA index number (1 to maximally 5) and type
  group_by(sgRNA_target) %>%
  mutate(
    sgRNA_position = as.numeric(sgRNA_position),
    sgRNA_index = sgRNA_position %>% as.factor %>% as.numeric,
    sgRNA_type = if_else(grepl("^nc_", sgRNA), "ncRNA", "gene")) %>%
  ungroup %>%
  
  # map trivial names to LocusTags using a manually curated list
  left_join(
    read_tsv("../data/input/mapping_trivial_names.tsv", col_types = cols()),
    by = c("sgRNA_target" = "gene")) %>%
  
  # remove some empty rows (NA targets)
  filter(!is.na(sgRNA_target)) %>%
  
  # remove 2 conditions without response
  filter(!condition %in% c("BG11", "LC, 200uE")) %>%
  
  # split condition into separate cols
  separate(condition, into = c("carbon", "light", "treatment_1", "treatment_2"),
    sep = ", ", remove = FALSE, fill = "right") %>%
  unite("treatment", treatment_1, treatment_2, sep = ", ", na.rm = TRUE)

Overview about the different conditions.

df_cultivation_summary <- df_main %>% group_by(condition) %>%
  summarize(
    time_points = paste(unique(time), collapse = ", "),
    carbon = unique(carbon),
    light = unique(light),
    treatment = unique(treatment),
    min_fit = min(fitness),
    med_fit = median(fitness),
    max_fit = max(fitness))

print(df_cultivation_summary)

write_csv(df_cultivation_summary, file = "../data/output/cultivation_summary.csv")

Retrieve gene info from uniprot and merge with main data frame. We need to make a custom function to retrieve and parse the data from uniprot, because of a bug in the security level on Ubuntu 20.04. The fallback option is to load a local copy of uniprot annotation for this organism.

library(httr)
uniprot_url <- paste0(
   "https://www.uniprot.org/uniprot/?query=taxonomy:1111708&format=tab&",
   "columns=id,genes,genes(PREFERRED),protein_names,length,mass,ec,database(KEGG)")

get_uniprot <- function(url) {
  # reset security level, caused by a faulty SSL certificate on server side,
  # see this thread: https://github.com/Ensembl/ensembl-rest/issues/427
  httr_config <- config(ssl_cipher_list = "DEFAULT@SECLEVEL=1")
  res <- with_config(config = httr_config, GET(url))
  server_error = simpleError("")
  df_uniprot <- tryCatch(
    read_tsv(content(res), col_types = cols()),
    error = function(server_error) {
      message("Uniprot server not available, falling back on local Uniprot DB copy")
      read_tsv("../data/input/uniprot_synechocystis.tsv", col_types = cols())
    }
  )
}

df_uniprot <- get_uniprot(uniprot_url) %>%
  rename_with(tolower) %>%
  rename(locus = `cross-reference (kegg)`, gene_name = `gene names`,
    gene_name_short = `gene names  (primary )`, ec_number = `ec number`,
    protein = `protein names`, uniprot_ID = entry
  ) %>%
  separate_rows(locus, sep = ";syn:") %>%
  mutate(locus = str_remove_all(locus, "syn:|;")) %>%
  filter(!is.na(locus))

df_main <- left_join(df_main, filter(df_uniprot, !duplicated(locus)),
  by = "locus")

3.3 Number of sgRNAs

Each gene is represented by up to five sgRNAs. We can test if all or only some of the 5 sgRNAs are “behaving” in the same way in the same conditions, more mathematically speaking we can estimate the correlation of every sgRNA with another. First let’s summarize how many genes have 5, 4, 3 sgRNAs and so on associated with them.

# N unique sgRNAs in dataset
paste0("Number of unique sgRNAs: ", unique(df_main$sgRNA) %>% length)

[1] "Number of unique sgRNAs: 21705"

# N genes with 1,2,3,4 or 5 sgRNAs
plot_sgRNAs_per_gene <- df_main %>%
  group_by(sgRNA_type, sgRNA_target) %>%
  summarize(n_sgRNAs = length(unique(sgRNA_position)), .groups = "drop_last") %>%
  count(n_sgRNAs) %>% filter(n_sgRNAs <= 5) %>%
  ggplot(aes(x = factor(n_sgRNAs, 5:1), y = n, label = n)) +
  geom_col(show.legend = FALSE) +
  geom_text(size = 3, nudge_y = 200, color = grey(0.5)) +
  facet_grid(~ sgRNA_type) +
  labs(x = "n sgRNAs / target", y = "n targets") +
  coord_cartesian(ylim = c(-50, 3500)) +
  custom_theme()

print(plot_sgRNAs_per_gene)

save_plot(plot_sgRNAs_per_gene, width = 6, height = 3.5)

4 Normalization

4.1 Fitness distribution of all conditions

Before biological analysis continues, we need to check if fitness (and log2 FC from which it is calculated) is equally distributed. For example, strictly essential genes like ribosomal genes should show the same degreee of depletion over time, regardless of condition.

We can compare fitness over all conditions using a scatter plot matrix. We can see that some conditions are very similar to each other, for example the conditions treated with glucose (LC, LL +g, LC, LL, +D, +G, HC, LL +g). Others are more dissimilar to the rest, for example LC, IL and LC, LL, +FL. They are more alike each other, although LC, LL, +FL should be more comparable to LC, LL, hinting at experimental bias. In this case both of these conditions (and LC, LL, +G) were pre-cultivated in low light instead of high light, as opposed to the rest of the samples.

df_main %>% filter(time == 0, sgRNA_index == 1) %>%
  select(locus, condition, fitness) %>%
  filter(!is.na(locus)) %>%
  pivot_wider(names_from = condition, values_from = fitness) %>%
  select(-locus) %>%
  custom_splom(pch = 19, cex = 0.3, col = grey(0.4, 0.4), pscales = 0)

4.2 Normalization strategy

In order to account for experimental or quantification bias, we can try to normalize the log2 FC distribution between all samples, and then re-calculate fitness. The underlying assumption is that e.g. essential genes should deplete at the same rate and hence show identical log2 FC at identical time points. Different types of experimental bias influence global fitness distribution and should be reduced with normalization. Here we try a ‘cyclic loess’ or quantile normalization that gave good results in a quick comparison.

# construct a normalization function that takes three colums as input,
# the numeric variable to be normalized, the conditioning variable
# (character or factor), and an ID that identifies each observation (sgRNA)
apply_norm = function(id, cond, var) {
  df_orig <- tibble(id = id, cond = cond, var = var)
  df_new <- pivot_wider(df_orig, names_from = cond, values_from = var) %>%
  column_to_rownames("id") %>% as.matrix %>%
  limma::normalizeBetweenArrays(method = "quantile") %>%
  as_tibble(rownames = "id") %>%
  pivot_longer(-id, names_to = "cond", values_to = "var_norm")
  left_join(df_orig, df_new, by = c("id", "cond")) %>% pull(var_norm)
}

# apply normalization
df_main <- df_main %>%
  mutate(FoldChange = 2^log2FoldChange) %>%
  group_by(time) %>%
  mutate(
    FoldChange_norm = apply_norm(sgRNA, condition, FoldChange),
    log2FoldChange = log2(FoldChange_norm)
  ) %>% ungroup

# compare effect of normalization
df_main %>% group_by(condition) %>% slice(1:10000) %>%
  ggplot(aes(x = log2(FoldChange), y = log2(FoldChange_norm), color = factor(time))) +
  geom_point(size = 0.5) +
  facet_wrap(~ condition, ncol = 4) +
  custom_theme() +
  scale_color_manual(values = custom_colors)

Another way to look at the result of the normalization is to compare the global distribution of log2 FC values, as a density plot.

library(ggridges)
df_main %>% filter(time == 10) %>%
  select(sgRNA, condition, FoldChange, FoldChange_norm) %>%
  pivot_longer(matches("^Fold"), names_to = "metric", values_to = "FC") %>%
  distinct %>%
  ggplot(aes(x = log2(FC), y = condition, group = condition)) + 
  geom_density_ridges(fill = "#00AFBB99", col = grey(0.4)) +
  facet_wrap(~ metric, ncol = 4) +
  lims(x = c(-2, 1.5)) +
  custom_theme()

Picking joint bandwidth of 0.0306
Picking joint bandwidth of 0.0312

Now we need to re-calculate fitness based on the normalized log2 FC.

df_main <- df_main %>%
  select(-FoldChange, -FoldChange_norm) %>%
  group_by(sgRNA, condition) %>%
  mutate(fitness = DescTools::AUC(time, log2FoldChange)/(max(time)/2)) %>%
  arrange(sgRNA_target, sgRNA_index, condition, time)

Registered S3 method overwritten by 'data.table':
  method           from
  print.data.table     

5 Fitness score aggregation

5.1 Correlation of sgRNAs

Different methods can be used to estimate similarity between samples (sgRNAs). For example, factor analysis is a method to dissect underlying sources of variation within the dataset, and the contribution to overall variation. The most famous example is principal component analysis (PCA). We can also use the correlation coefficient of sgRNAs to each other to see if one of the sgRNAs contributes stronger to overall variation.

This is an example of an apparently strictly essential gene, encoding the ribosomal protein rps10. Most of the sgRNA repressor strains are depleted, the correlation between sgRNAs is high. The strength of depletion varies though, and the strain with sgRNA 3 is not depleted at all. We want to give higher weights to sgRNAs that correlate well with each other, and/or show stronger effect (depletion/enrichment).

plot_sgRNA_ribo_example <- df_main %>% filter(sgRNA_target == "rps10") %>%
  mutate(sgRNA_index = factor(sgRNA_index, 1:5)) %>%
  ggplot(aes(x = time, y = log2FoldChange, color = sgRNA_index)) +
  geom_line(size = 1) + geom_point(size = 2) +
  facet_wrap(~ condition, ncol = 4) +
  custom_theme() +
  scale_color_manual(values = custom_range(5))

print(plot_sgRNA_ribo_example)

save_plot(plot_sgRNA_ribo_example, width = 7, height = 5.5)

A correlation score can be calculated by computing the correlation coefficient of all sgRNAs to each other. This score is robustly summarized by taking the median, and rescaling it from the respective minima and maxima [-1, 1] to [0, 1]. This score serves as a weight component for each sgRNA to calculate the (global) weighted mean of log2 FC over all sgRNAs. The score has the characteristic that it gives a weight of 1 for an sgRNA perfectly correlated with all other sgRNAs of the same gene, and a weight of 0 for sgRNAs perfectly anti-correlated to the other sgRNAs.

For a matrix of \(x = 1 .. m\) sgRNAs and \(y = 1 .. n\) observations (measurements), the correlation \(R\) of one sgRNA to another is calculated using Pearson’s method:

\(R_x=cor([log_2FC_{x1,y1} ... log_2FC_{x1,yn}], [log_2FC_{x2,y1} ... log_2FC_{x2,yn}])\)

The correlation weight of one sgRNA is then calculated as median of all \(R\) rescaled between 0 and 1.

\(w_x = \frac{1 + median(R_1, R_2, ..., R_m)}{2}\)

The following example shows the correlation matrix for the 5 rps10 sgRNAs, and their weights. The self correlation of each sgRNA (R = 1) is removed prior to weight determination.

cor_matrix <- df_main %>% filter(sgRNA_target == "rps10") %>% ungroup %>%
  select(sgRNA_index, log2FoldChange, condition, time) %>%
  pivot_wider(names_from = c("condition", "time"), values_from = log2FoldChange) %>%
  arrange(sgRNA_index) %>% column_to_rownames("sgRNA_index") %>%
  as.matrix %>% t %>% cor(method = "pearson")

weights <- cor_matrix %>% replace(., . == 1, NA) %>%
  apply(2, function(x) median(x, na.rm = TRUE)) %>%
  rescale(from = c(-1, 1), to = c(0, 1))

# plot heatmap
lattice::levelplot(cor_matrix %>% replace(., . == 1, NA),
  col.regions = custom_range(20))

# print weights
weights

        1         2         3         4         5 
0.8440521 0.7864564 0.4605635 0.8265134 0.7689177 

---

Now we can create a function that will compute weights for all sgRNAs, and add the weights to the data set.

determine_corr <- function(index, value, condition, time) {
  # make correlation matrix
  df <- data.frame(index = index, value = value, condition = condition, time = time)
  cor_matrix <- pivot_wider(df, names_from = c("condition", "time"), values_from = value) %>%
    arrange(index) %>% column_to_rownames("index") %>%
    as.matrix %>% t %>% cor(method = "pearson")
  
  # determine weights
  weights <- cor_matrix %>% replace(., . == 1, NA) %>%
    apply(2, function(x) median(x, na.rm = TRUE)) %>%
    scales::rescale(from = c(-1, 1), to = c(0, 1)) %>%
    enframe("index", "weight") %>% mutate(index = as.numeric(index)) %>%
    mutate(weight = replace(weight, is.na(weight), 1))
  
  # return vector of weights the same order and length 
  # as sgRNA index vector
  left_join(df, weights, by = "index") %>% pull(weight)
}

df_main <- df_main %>%
  group_by(sgRNA_target) %>%
  mutate(sgRNA_correlation = determine_corr(sgRNA_index,
    log2FoldChange, condition, time))

5.2 Efficiency of sgRNAs

The correlation of each sgRNA with each other is a “global” parameter as it is identical over all conditions. A second global parameter, sgRNA efficiency, can be obtained using a similar approach. We expect that fitness of all sgRNAs for one gene is not normally distributed because sgRNAs are not ideal replicate measurements. They are biased by position effects and off-target binding, see Wang et al., Nature Comms, 2018 for a very insightful and comprehensive analysis of the number and position of sgRNAs required to estimate gene fitness.

We calculate sgRNA efficiency \(E\) as the median absolute fitness (AUC of log2FC over time) of an sgRNA \(x = 1 .. m\) over all observations [conditions] \(y = 1 .. n\).

\(E_x=median(abs(fitness_{x1, y1}, fitness_{x1, y2}, ..., fitness_{x1, yn}))\)

To normalize between all sgRNAs, \(E\) is rescaled to a range between 0 and 1.

\(E_x=\frac{E_x}{max(E_1, E_2, ..., E_m)}\)

df_main <- df_main %>% group_by(sgRNA_target) %>%
  mutate(sgRNA_efficiency = ave(fitness, sgRNA_index, FUN = function(x) median(abs(x))) %>%
    {./max(.)})

This is the resulting sgRNA efficiency for the example gene above, rps10.

df_main %>% filter(sgRNA_target == "rps10") %>% ungroup %>%
  select(sgRNA_index, sgRNA_efficiency) %>% distinct %>% 
  arrange(sgRNA_index) %>% deframe

        1         2         3         4         5 
1.0000000 0.1519365 0.0351794 0.2105323 0.5110918 

5.3 Position bias of sgRNA repression

Plot the weight of each sgRNA to see if there is a dependency between correlation and sgRNA position. There is no significant trend.

We can also quantify how many genes have strongly correlated sgRNAs and how many have outliers. In order to do this, the median weight of the (up to) 5 sgRNAs per gene is plotted. Generally, the median weight ranges between 0.5 and 1.0, showing on average good correlation.

plot_sgRNA_correlation <- df_main %>%
  select(sgRNA_target, sgRNA_index, sgRNA_correlation) %>%
  filter(sgRNA_index <= 5) %>%
  distinct %>%
  # plot
  ggplot(aes(x = factor(sgRNA_index), y = sgRNA_correlation)) +
  geom_boxplot(outlier.shape = "") +
  labs(x = "sgRNA position", y = "correlation") +
  stat_summary(fun.data = function(x) c(y = median(x)+0.07, 
    label = round(median(x), 2)), geom = "text", size = 3) +
  stat_summary(fun.data = function(x) c(y = 1.1, 
    label = length(x)), geom = "text", color = grey(0.5), size = 3) +
  coord_cartesian(ylim = c(-0.15, 1.15)) +
  custom_theme()

plot_sgRNA_correlation_hist <- df_main %>%
  select(sgRNA_target, sgRNA_index, sgRNA_correlation) %>%
  filter(sgRNA_index <= 5) %>%
  distinct %>% group_by(sgRNA_target) %>%
  summarize(
    median_sgRNA_correlation = median(sgRNA_correlation),
    min_sgRNA_correlation = min(sgRNA_correlation)
  ) %>%
  # plot
  ggplot(aes(x = median_sgRNA_correlation)) +
  geom_histogram(bins = 40, fill = custom_colors[1], alpha = 0.7) +
  custom_theme()

save_plot(plot_sgRNA_correlation_hist, width = 5, height = 4)
save_plot(plot_sgRNA_correlation, width = 5, height = 4)
ggarrange(plot_sgRNA_correlation, plot_sgRNA_correlation_hist, ncol = 2)

Second, the binding position of the sgRNAs could be correlated to the strength of repression. In other words sgRNAs binding closer to the promoter could have stronger ability to repress a gene, see Figure 1 B in Wang et al., Nature Comms, 2018. We plot sgRNA efficiency for genes only, because the absolute majority of those has 5 sgRNAs.

plot_sgRNA_efficiency <- df_main %>%
  filter(sgRNA_index <= 5, sgRNA_type == "gene") %>%
  select(sgRNA_target, sgRNA_index, sgRNA_efficiency) %>% distinct %>%
  ggplot(aes(x = factor(sgRNA_index), y = sgRNA_efficiency)) +
  geom_boxplot(notch = FALSE, outlier.shape = ".") +
  labs(x = "sgRNA position (relative)", y = "repression efficiency") +
  coord_cartesian(ylim = c(-0.15, 1.15)) +
  stat_summary(fun.data = function(x) c(y = median(x)+0.07, 
    label = round(median(x), 2)), geom = "text", size = 3) +
  stat_summary(fun.data = function(x) c(y = 1.1, 
    label = length(x)), geom = "text", color = grey(0.5), size = 3) +
  custom_theme()


plot_sgRNA_efficiency_hist <- df_main %>%
  filter(sgRNA_index <= 5, sgRNA_type == "gene") %>%
  select(sgRNA_target, sgRNA_position, sgRNA_efficiency) %>% distinct %>%
  group_by(sgRNA_position) %>%
  summarize(sgRNA_efficiency = median(sgRNA_efficiency), n_pos = n()) %>%
  filter(n_pos >= 10) %>%
  ggplot(aes(x = sgRNA_position, y = sgRNA_efficiency)) +
  labs(x = "sgRNA position (nt)", y = "repression efficiency") +
  geom_point(col = alpha(custom_colors[5], 0.5)) +
  geom_smooth() +
  custom_theme()

save_plot(plot_sgRNA_efficiency, width = 5, height = 4)
save_plot(plot_sgRNA_efficiency_hist, width = 5, height = 4)

`geom_smooth()` using method = 'loess' and formula 'y ~ x'
`geom_smooth()` using method = 'loess' and formula 'y ~ x'

ggarrange(plot_sgRNA_efficiency, plot_sgRNA_efficiency_hist, ncol = 2)

`geom_smooth()` using method = 'loess' and formula 'y ~ x'

Export draft Figure 1 for manuscript.

plot_selected_sgRNAs <- df_main %>%
  filter(
    grepl("ctrl[1-5]$|rps10$", sgRNA_target), 
    condition %in% c("HC, HL", "HC, LL", "LC, IL", "LC, LL")) %>%
  mutate(
    sgRNA_index2 = as.numeric(str_extract(sgRNA_target, "[1-9]$")),
    sgRNA_index = case_when(sgRNA_position == 0 ~ sgRNA_index2, TRUE ~ sgRNA_index),
    sgRNA_target = str_extract(sgRNA_target, "[a-zA-Z]*")
  ) %>%
  ggplot(aes(x = time, y = log2FoldChange, color = factor(sgRNA_index))) +
  geom_line(size = 1) + geom_point(size = 2) +
  facet_grid(sgRNA_target ~ condition) +
  custom_theme(legend.position = 0) +
  coord_cartesian(ylim = c(-4.5, 2.5)) +
  scale_color_manual(values = custom_range(5))

svg(filename = "../figures/figure1.svg", width = 7, height = 5.5)
ggarrange(ncol = 2, nrow = 2, widths = c(0.6, 0.4), labels = LETTERS[1:4], font.label = list_fontpars,
  plot_sgRNAs_per_gene + theme(plot.margin = unit(c(12,12,12,12), "points")),
  plot_sgRNA_efficiency + theme(plot.margin = unit(c(26,12,12,12), "points")),
  plot_selected_sgRNAs + theme(plot.margin = unit(c(12,-4,12,14), "points")),
  plot_sgRNA_correlation + theme(plot.margin = unit(c(26,12,12,12), "points"))
)
dev.off()

null device 
          1 

Export supplemental figure with all ribosomal genes (rpsNN/rplNN).

plot_sgRNAs_ribosome <- df_main %>%
  filter(str_detect(sgRNA_target, "rp[sl][0-9]*$")) %>%
  filter(condition == "LC, LL") %>%
  ggplot(aes(x = time, y = log2FoldChange, color = factor(sgRNA_index))) +
  geom_line(size = 1) + geom_point(size = 2) +
  facet_wrap(~ sgRNA_target, ncol = 7) +
  custom_theme(legend.position = "top") +
  scale_color_manual(values = custom_range(5))

print(plot_sgRNAs_ribosome)

6 Gene fitness calculation

6.1 Summarize sgRNA fitness to gene fitness

With the correlation and the efficiency per sgRNA, we can compute the weighted mean of all sgRNAs. For comparison, we also test simple strategies such as the standard arithmetic mean and a top 1 and top 2 sgRNAs strategy. Metrics are calculated for log2 FC, and fitness.

df_controls <- df_main %>% ungroup %>% 
  filter(str_detect(sgRNA_target, "ctrl[0-9]+$"))

df_gene <- df_main %>%
  
  # keep all annotation columns
  group_by(sgRNA_target, sgRNA_type, locus, gene_name, condition, 
    carbon, light, treatment, time) %>%
  
  # summarize FC and fitness...
  summarize(.groups = "drop",
    
    # log2 FC
    mean_log2FoldChange = mean(log2FoldChange),
    wmean_log2FoldChange = weighted.mean(log2FoldChange, sgRNA_correlation * sgRNA_efficiency),
    top1_log2FoldChange = log2FoldChange[which.max(sgRNA_efficiency)],
    top2_log2FoldChange = mean(log2FoldChange[order(sgRNA_efficiency, decreasing = TRUE)[1:2]]),
    sd_log2FoldChange = sd(log2FoldChange),
    
    # fitness
    mean_fitness = mean(fitness),
    wmean_fitness = weighted.mean(fitness, sgRNA_correlation * sgRNA_efficiency),
    top1_fitness = fitness[which.max(sgRNA_efficiency)],
    top2_fitness = mean(fitness[order(sgRNA_efficiency, decreasing = TRUE)[1:2]]),
    sd_fitness = sd(fitness),
    
    # apply significance test, Mann-Whitney U test
    p_value = wilcox.test(fitness, filter(df_controls, condition == unique(condition))$fitness)$p.value
  )

Since statistical significance is tested for many genes in parallel, the p-value obtained from MWU test should be multiple-hypothesis corrected. For this purpose we use the Benjamini-Hochberg method. We also calculate a score taking both effect size and p-value into account, according to the publication from Wang et al., Nat Comm, 2018. This score is simply the absolute fitness score multiplied by the negative log10 p-value.

df_gene <- df_gene %>%
  group_by(condition, time) %>%
  mutate(
    p_value_adj = p.adjust(p_value, method = "BH"),
    score = abs(wmean_fitness)*-log10(p_value_adj)
  ) %>% ungroup

A comparison of log2 FC aggregated by the different method shows clear differences. For the example gene rps10 the weighted mean and the top method give similar results, representative of the stronger influence from highly depleted sgRNA repression strains. The regular mean is robust, but “shallow”, probably underestimating the real effect n fitness. The top 1 method simply picks the most depleted/enriched sgRNA (over all conditions) as representative.

df_gene %>% filter(sgRNA_target == "rps10") %>%
  pivot_longer(cols = matches("[n12]_log2FoldChange"), 
    names_to = "metric", values_to = "log2FoldChange") %>%
  mutate(metric = str_remove(metric, "_log2FoldChange")) %>%
  ggplot(aes(x = time, y = log2FoldChange, 
    ymin = log2FoldChange-sd_log2FoldChange, 
    ymax = log2FoldChange+sd_log2FoldChange, color = fct_inorder(metric))) +
  geom_line(size = 1) + geom_point(size = 2) + geom_linerange(size = 1) +
  facet_wrap(~ condition, ncol = 4) +
  custom_theme(legend.position = "right") +
  coord_cartesian(ylim = c(-3.75, 0.75)) +
  scale_color_manual(values = custom_range(4))

This plot shows a comparison of the 4 methods for the first 36 genes by alphabetical order, for one selected condition only (1% CO2, BG11, 1,000 µmol photons m-1 s-1). Here we can see that the top1 method is often but not always representative for the gene: For apcD or apcF, it does not seem representative compared to the mean, weighted mean, and top2 methods.

df_gene %>% filter(
    gene_name %in% unique(.data[["gene_name"]])[1:36],
    condition == "HC, HL"
  ) %>%
  pivot_longer(cols = matches("[n12]_log2FoldChange"), names_to = "metric", values_to = "log2FoldChange") %>%
  mutate(metric = str_remove(metric, "_log2FoldChange")) %>%
  ggplot(aes(x = time, y = log2FoldChange, 
    ymin = log2FoldChange-sd_log2FoldChange,
    ymax = log2FoldChange+sd_log2FoldChange, color = fct_inorder(metric))) +
  geom_line(size = 1) + geom_point(size = 2) + geom_linerange(size = 1) +
  facet_wrap(~ sgRNA_target, ncol = 7) +
  custom_theme(legend.position = "top") +
  coord_cartesian(ylim = c(-5, 5)) +
  scale_color_manual(values = custom_range(4))

6.2 Global distribution of gene fitness

Global distribution of weighted mean fitness for all genes. Effect of ncRNA repression seems to be much lower than effect of gene repression.

plot_all_fitness_hist <- df_gene %>% filter(time == 0) %>%
  ggplot(aes(x = wmean_fitness, fill = sgRNA_type)) +
  geom_histogram(bins = 100) +
  coord_cartesian(xlim = c(-4, 4), ylim = c(0, 1000)) +
  facet_wrap( ~ condition, ncol = 4) +
  custom_theme() +
  scale_fill_manual(values = custom_colors[c(3:4)])

print(plot_all_fitness_hist)

save_plot(plot_all_fitness_hist, width = 7, height = 5)

6.3 Gene fitness vs significance

plot_all_fitness_volc <- df_gene %>% filter(time == 0,
    condition %in% c("HC, HL", "LC, LL")) %>%
  arrange(sgRNA_type) %>%
  ggplot(aes(x = wmean_fitness, y = -log10(p_value_adj), col = sgRNA_type)) +
  geom_point(alpha = 0.5, size = 0.5) +
  geom_line(data = data.frame(x = c(seq(-6, -0.5, 0.1), seq(0.5, 6, 0.1)),
    y = 4/c(seq(6, 0.5, -0.1), seq(0.5, 6, 0.1))),
    aes(x = x, y = y, shape = NULL, col = NULL), lty = 2) +
  coord_cartesian(xlim = c(-7, 7), ylim = c(0, 4)) +
  custom_theme(aspect = 1, legend.position = "left", legend.key.size = unit(0.4, "cm")) +
  facet_wrap(~ condition) +
  labs(x = "fitness", y = expression("-log"[10]*" p-value")) +
  scale_color_manual(values = custom_colors[3:4]) +
  scale_shape_manual(values=c(1, 19))

print(plot_all_fitness_volc)

save_plot(plot_all_fitness_volc, width = 6, height = 3)

6.4 Behavior of control sgRNAs

Ten sgRNAs were included in the library that have no gene-specific targets. The following plot shows that these negative controls do not have an effect on strain fitness, except probably 2 sgRNAs in one specific condition.

plot_controls_sgRNAs <- df_main %>% filter(grepl("ctrl", sgRNA_target)) %>%
  ggplot(aes(x = time, y = log2FoldChange, color = sgRNA_target)) +
  geom_line(size = 1) + geom_point(size = 2) + ylim(-5, 5) +
  facet_wrap(~ condition, ncol = 4) +
  custom_theme() +
  scale_color_manual(values = custom_range(10))

print(plot_controls_sgRNAs)

save_plot(plot_controls_sgRNAs, width = 7, height = 5.5)

7 Gene enrichment

To plot gene fitness for the enzymes of central carbon metabolism, we need a complete list of enzymes and the genes that they are mapped to. To list the different KEGG databases that can be queried, use listDatabases(). Gene-pathway mappings are obtained and merged with pathway names and gene/enzyme names.

# get mapping of pathways for each gene
df_kegg <- keggLink("pathway", "syn") %>%
  enframe(name = "locus", value = "kegg_pathway_id") %>%
  
  # get list of pathways with name/ID pairs
  left_join(by = "kegg_pathway_id",
    keggList("pathway", "syn") %>%
    enframe(name = "kegg_pathway_id", value = "kegg_pathway")
  ) %>%
  
  # get list of gene/enzyme names
  left_join(by = "locus",
    keggList("syn") %>%
    enframe(name = "locus", value = "kegg_gene") %>%
    mutate(kegg_gene_short = str_extract(kegg_gene, "^[a-zA-Z0-9]*;") %>% 
      str_remove(";"))
  ) %>%
  
  # trim useless prefixes
  mutate(
    locus = str_remove(locus, "syn:"),
    kegg_pathway_id = str_remove(kegg_pathway_id, "path:"),
    kegg_pathway = str_remove(kegg_pathway, " - Synechocystis sp. PCC 6803")
  )

head(df_kegg)

7.1 Fitness per pathway

Sometimes even small effects in fitness can be relevant if several genes of the same pathway (or iso-enzymes) are affected. A simple fitness threshold will not reveal those changes. In such cases a more nuanced approach can be taken, a gene set enrichment analysis (GSEA). Several packages exist to test if functionally related genes are enriched, depleted, or both at the same time / the same conditions.

Before we test for enrichment of associated pathways/GO terms, we can have a look at the general depletion/enrichment per KEGG pathway. The fitness distribution per pathway can be visualized using a violin- or scatter plot.

plot_median_fitness_kegg <- df_gene %>% filter(time == 0) %>%
  inner_join(df_kegg, by = "locus") %>%
  group_by(kegg_pathway, condition) %>%
  summarize(.groups = "drop",
    fitness = median(wmean_fitness),
    n_genes = n()
  ) %>% filter(n_genes >= 20) %>%
  mutate(kegg_pathway = paste0(str_sub(kegg_pathway, 1, 25), "..")) %>%
  mutate(kegg_pathway = fct_reorder(kegg_pathway, fitness, .desc = TRUE)) %>%
  
  ggplot(aes(x = fitness, y = kegg_pathway)) +
  geom_boxplot(outlier.shape = NULL, color = grey(0.5), fill = grey(0.9)) +
  geom_point(aes(color = condition)) +
  geom_vline(xintercept = 0, lty = 2, color = grey(0.5)) +
  labs(x = "median fitness", y = "") +
  custom_theme(legend.position = c(0.25, 0.25), legend.key.size = unit(0.4, "cm")) +
  scale_fill_manual(values = custom_range(11)) +
  scale_color_manual(values = custom_range(11))

print(plot_median_fitness_kegg)

Export draft Figure 2 for manuscript. We add photosystem I and II genes as examples for differential depletion. A heatmap.

plot_sgRNAs_ps1 <- df_gene %>%
  filter(str_detect(sgRNA_target, "psa[A-Z]*"), time == 0) %>%
  mutate(wmean_fitness = wmean_fitness %>% replace(., . > 4, 4) %>% replace(., . < -4, -4)) %>%
  ggplot(aes(x = condition, y = fct_rev(sgRNA_target), fill = wmean_fitness)) +
  geom_tile() + custom_theme() +
  labs(title = "Photosystem I", x = "", y = "") +
  theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1)) +
  scale_fill_gradientn(colours = c(custom_colors[1], grey(0.9), custom_colors[2]),
    limits = c(-4, 4))

plot_sgRNAs_ps2 <- df_gene %>%
  filter(str_detect(sgRNA_target, "psb[A-Z]*"), time == 0) %>%
  mutate(wmean_fitness = wmean_fitness %>% replace(., . > 4, 4) %>% replace(., . < -4, -4)) %>%
  mutate(sgRNA_target = str_replace(sgRNA_target, "psb13", "psbW")) %>%
  ggplot(aes(x = condition, y = fct_rev(sgRNA_target), fill = wmean_fitness)) +
  geom_tile() + custom_theme() +
  labs(title = "Photosystem II", x = "", y = "") +
  theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1)) +
  scale_fill_gradientn(colours = c(custom_colors[1], grey(0.9), custom_colors[2]),
    limits = c(-4, 4))

ggarrange(ncol = 2, plot_sgRNAs_ps1, plot_sgRNAs_ps2)

svg(filename = "../figures/figure2.svg", width = 8, height = 7)
ggarrange(ncol = 2, widths = c(0.65, 0.35),
  ggarrange(nrow = 2, heights =  c(0.34, 0.66), labels = LETTERS[1:2], font.label = list_fontpars,
    plot_all_fitness_volc + theme(plot.margin = unit(c(14,-8,14,40), "points")),
    plot_median_fitness_kegg + theme(plot.margin = unit(c(6,12,12,12), "points"))),
  ggarrange(nrow = 2, heights =  c(0.4, 0.6), labels = LETTERS[3:4], font.label = list_fontpars,
    plot_sgRNAs_ps1 + theme(plot.margin = unit(c(12,0,-14,0), "points")),
    plot_sgRNAs_ps2 + theme(plot.margin = unit(c(12,0,0,0), "points"))
  )
)
dev.off()

null device 
          1 

7.2 Gene enrichment analysis (KEGG)

We use the functions kegga for KEGG enrichment analysis and goana for GO term enrichment from the limma package. Both functions test for over or under-representation of genes associated with certain pathways or GO terms. The functions don’t take the strength of differential fitness into account (DF; the depletion/enrichment over time).

df_kegg_enrichment <- lapply(unique(df_gene$condition), function(cond) {
  df_gene %>% filter(
  sgRNA_type == "gene", time == 0,
  condition == cond) %>%
  
  # filter for differential fitness (DF) genes
  filter(!between(wmean_fitness, -2.0, 2.0), !is.na(locus)) %>%
  
  # perform KEGG enrichment
  pull(locus) %>% kegga(species.KEGG = "syn") %>%
  mutate(condition = cond)
}) %>% bind_rows

head(df_kegg_enrichment)

Now we visualize the pathways that are most enriched for DF genes. It turns out that ribosomal proteins are extremely depleted and therefore score high on the negative log10 p-value for pathway enrichment.

df_kegg_enrichment %>%
  rename(kegg_pathway = Pathway) %>%
  group_by(kegg_pathway) %>% filter(N >= 20) %>%
  select(kegg_pathway, condition, P.DE) %>%
  mutate(log10_p_value = -log10(P.DE), .keep = "unused") %>%
  mutate(kegg_pathway = paste0(str_sub(kegg_pathway, 1, 25), "..")) %>%
  
  # make correlation plot
  pivot_wider(names_from = condition, values_from = log10_p_value) %>%
  column_to_rownames(var = "kegg_pathway") %>% as.matrix %>%
  corrplot(is.corr = FALSE, tl.col = grey(0.5), tl.cex = 0.8,
    col = colorRampPalette(custom_colors[c(1,5,2)])(10), col.lim = c(0, 20))

8 Unsupervised clustering of genes

8.1 Cluster genes by similarity

# generate color palette for heatmap
heat_cols <- colorspace::diverging_hcl(n = 7, h = c(255, 12), c = c(50, 80), l = c(20, 97), power = c(1, 1.3))

# create a matrix-like df with wide fitness data for plotting heatmap
df_heatmap <- df_gene %>%
  filter(time == 0, !is.na(locus)) %>%
  select(locus, condition, wmean_fitness) %>%
  mutate(wmean_fitness = wmean_fitness %>% replace(., . > 8, 8) %>% replace(., . < -8, -8)) %>%
  pivot_wider(names_from = condition, values_from = wmean_fitness) %>%
  column_to_rownames(var = "locus")

# subset of df with *strongly changed* genes
df_heatmap2 <- df_heatmap %>%
  filter(if_any(.cols = matches("[HL]C, "), ~ !between(., -4, 4)))

# create cluster for reordering
mat_cluster <- df_heatmap %>% as.matrix %>% dist %>% hclust(method = "ward.D2")
mat_heatmap <- df_heatmap %>% as.matrix %>%
  .[order.dendrogram(as.dendrogram(mat_cluster)), ncol(.):1]

# repeat this step with subset of *significantly changed* genes
mat_cluster_sig <- df_heatmap2 %>% as.matrix %>% dist %>% hclust(method = "ward.D2")
mat_heatmap_sig <- df_heatmap2 %>% as.matrix %>%
  .[order.dendrogram(as.dendrogram(mat_cluster_sig)), ncol(.):1]

Now we can plot all genes, a subset with only significant genes, and a dendrogram for clustering. The result is hard to interpret. With some exceptions, most genes are grouped in broad unspecific clusters that do not reveal clear relationships between treatment variables and fitness outcome.

plot_heatmap_all <- levelplot(mat_heatmap,
  par.settings = custom.colorblind(),
  col.regions = colorRampPalette(heat_cols)(20),
  at = seq(-8, 8, 1), aspect = "fill",
  xlab = paste0("genes (", nrow(mat_heatmap),")"),
  ylab = "", scales = list(x = list(draw = FALSE)),
  panel = function(x, y, z, ...) {
    panel.levelplot(x, y, z, ...)
    panel.abline(h = 1:5+0.5, col = "white", lwd = 1.5)
  }
)

print(plot_heatmap_all)

save_plot(plot_heatmap_all, width = 8, height = 2.5)

plot_heatmap_sig <- levelplot(mat_heatmap_sig,
  par.settings = custom.colorblind(),
  col.regions = colorRampPalette(heat_cols)(20),
  at = seq(-8, 8, 1), aspect = "fill",
  xlab = paste0("genes (", nrow(mat_heatmap_sig),")"),
  ylab = "", scales = list(x = list(draw = FALSE)),
  panel = function(x, y, z, ...) {
    panel.levelplot(x, y, z, ...)
    panel.abline(h = 1:ncol(mat_heatmap_sig)+0.5, col = "white", lwd = 1.5)
  }
)

plot_cluster_dend <- mat_cluster_sig %>% as.dendrogram %>%
  set("branches_k_col", custom_colors[1:5], k = 5) %>%
  set("branches_lwd", 0.5) %>%
  as.ggdend %>%
  ggplot(labels = FALSE)

gridExtra::grid.arrange(
  # coords for unit: top, right, bottom, left
  plot_cluster_dend + 
    theme(plot.margin = unit(c(0.1, 0.075, -0.26, 0.135),"npc")),
  plot_heatmap_sig,
  nrow = 2
)

8.2 Gene similarity by dimensionality reduction methods

We use two different dimensionality reduction methods, nMDS and t-SNE. We can check if these methods reproduce the clustering for the significantly regulated genes produced with hclust. Analysis shows that the small clusters are more strongly separated from the rest.

# set a seed to obtain same pattern for stochastic methods
set.seed(123)

# run nMDS analysis
NMDS <- df_heatmap2 %>% as.matrix %>% dist %>% metaMDS

Run 0 stress 0.08477877 
Run 1 stress 0.08485663 
... Procrustes: rmse 0.002551692  max resid 0.04275329 
Run 2 stress 0.08484567 
... Procrustes: rmse 0.001445249  max resid 0.01754732 
Run 3 stress 0.09041578 
Run 4 stress 0.09773472 
Run 5 stress 0.09766743 
Run 6 stress 0.09766712 
Run 7 stress 0.09766695 
Run 8 stress 0.097774 
Run 9 stress 0.08581093 
Run 10 stress 0.08477915 
... Procrustes: rmse 0.0001249276  max resid 0.00204171 
... Similar to previous best
Run 11 stress 0.09034496 
Run 12 stress 0.08477945 
... Procrustes: rmse 0.0001738428  max resid 0.002843804 
... Similar to previous best
Run 13 stress 0.08477977 
... Procrustes: rmse 0.0002175779  max resid 0.003564318 
... Similar to previous best
Run 14 stress 0.08477903 
... Procrustes: rmse 0.0001075038  max resid 0.001756538 
... Similar to previous best
Run 15 stress 0.08484495 
... Procrustes: rmse 0.001423391  max resid 0.01739739 
Run 16 stress 0.08485667 
... Procrustes: rmse 0.002551594  max resid 0.0427536 
Run 17 stress 0.08485605 
... Procrustes: rmse 0.002540429  max resid 0.04264689 
Run 18 stress 0.08477984 
... Procrustes: rmse 0.0002060804  max resid 0.003370265 
... Similar to previous best
Run 19 stress 0.09041548 
Run 20 stress 0.09286828 
*** Solution reached

df_nmds <- NMDS$points %>% as_tibble(rownames = "locus") %>%
  left_join(enframe(cutreeord(mat_cluster_sig, k = 5), "locus", "cluster"))

Joining, by = "locus"

# run t-SNE analysis
SNE <- tsne::tsne(df_heatmap2 %>% as.matrix %>% dist)

sigma summary: Min. : 0.51453154858122 |1st Qu. : 0.646910330520664 |Median : 0.714309515627356 |Mean : 0.746309912193634 |3rd Qu. : 0.79479345979067 |Max. : 1.58041817397143 |
Epoch: Iteration #100 error is: 14.170258824269
Epoch: Iteration #200 error is: 0.423942816192504
Epoch: Iteration #300 error is: 0.406398188540306
Epoch: Iteration #400 error is: 0.404698129239104
Epoch: Iteration #500 error is: 0.404337884848916
Epoch: Iteration #600 error is: 0.404205045213674
Epoch: Iteration #700 error is: 0.404146174344848
Epoch: Iteration #800 error is: 0.404117696326219
Epoch: Iteration #900 error is: 0.404102464318572
Epoch: Iteration #1000 error is: 0.404093843400558

df_tsne <- SNE %>% setNames(c("x", "y")) %>% as_tibble %>%
  mutate(locus = rownames(df_heatmap2)) %>%
  left_join(enframe(cutreeord(mat_cluster_sig, k = 5), "locus", "cluster"))

Joining, by = "locus"

plot_nmds <- df_nmds %>% 
  ggplot(aes(x = MDS1, y = MDS2, color = factor(cluster))) +
  geom_point(size = 2) + labs(title = "nMDS") +
  custom_theme(legend.position = c(0.85, 0.78)) +
  scale_color_manual(values = custom_colors)

plot_tsne <- df_tsne %>% 
  ggplot(aes(x = V1, y = V2, color = factor(cluster))) +
  geom_point(size = 2) + labs(title = "t-SNE") +
  custom_theme(legend.position = c(0.85, 0.78)) +
  scale_color_manual(values = custom_colors)

gridExtra::grid.arrange(plot_nmds, plot_tsne, ncol = 2)

ggsave("../figures/plot_nmds_tsne.svg",
  plot = gridExtra::arrangeGrob(plot_nmds, plot_tsne, ncol = 2),
  device = "svg", width = 8, height = 4)

8.3 Fit multiple linear regression models

We can find clusters of genes with similar fitness, but it is also important to identify why they cluster together. In order to find out which variables determine the fitness outcome of a gene, we can perform multiple linear regression. Each gene needs to have fitness outcomes annotated with the different (mixed) variables carbon, light, treatment. The latter can be subdivided in individual treatment columns glucose, DCMU, fluctuating light, and so on. Multiple linear regression fits a linear model of the following form to the data:

response ~ intercept + predictor A x slope A + predictor B x slope B x ...

Here, fitness is the response variable, the different conditions are the predictors. It is important to convert the categorical predictors into (numerical) dummy variables. Then for each individual gene, multiple linear models are fitted and the power of each predictor variable to predict the response is extracted.

# fixed model with 6 predictor variables -- dynamic layout would 
# be better in future
fit_linreg <- function(y, x1, x2, x3, x4, x5, x6){
  fit <- lm(y ~ x1 + x2 + x3 + x4 + x5 + x6)
  c(coefficients(fit), summary(fit)$coefficients[, 4],
    summary(fit)$r.squared)
}

# recode categorical to numerical (dummy) variables
df_linreg <- df_gene %>%
  filter(!is.na(locus)) %>%
  select(locus, carbon, light, treatment, wmean_fitness) %>% distinct %>%
  mutate(
    carbon = recode(carbon, `HC` = 1, `LC` = 0),
    light = recode(light, `LL` = 0, `IL` = 0.5, `HL` = 1)) %>%
  mutate(dummy = 1, treatment = replace(treatment, treatment == "", "-")) %>%
  pivot_wider(names_from = treatment, values_from = dummy, values_fill = 0) %>%
  mutate(`+G` = `+G` + `+D, +G`) %>% rename(`+D` = `+D, +G`) %>% select(-`-`) %>%
  # fit model
  group_by(locus) %>%
  summarize(coefficient = fit_linreg(wmean_fitness, carbon, light, `-N`, `+FL`, `+G`, `+D`),
    .groups = "keep") %>% #unnest(coefficient) %>%
  mutate(treatment = c(rep(c("intercept", "carbon", "light", "-N", "+FL", "+G", "+D"), 2) %>% 
    paste0(rep(c("", "pval_"), each = 7), .), "r_squared"))

Now we can overlay the information of the best predictor variable on the cluster map produced by tSNE, for example, and this way identify groups of genes regulated in a similar degree, by similar variables.

plot_tsne_linreg <- df_tsne %>%
  inner_join(df_linreg, by = "locus") %>%
  left_join(select(df_gene, locus, sgRNA_target) %>% distinct, by = "locus") %>%
  filter(!str_detect(treatment, "intercept|pval|r_squared")) %>%
  mutate(sgRNA_target = if_else(abs(coefficient) > 2, sgRNA_target, "")) %>%
  mutate(point_size = abs(coefficient),
    coefficient = coefficient %>% replace(., . > 5, 5) %>% replace(., . < -5, -5)) %>%
  
  ggplot(aes(x = V1, y = V2, size = point_size,
    color = coefficient, label = sgRNA_target)) +
  geom_point() +
  labs(title = "t-SNE clustering of DF genes", 
    subtitle = paste0("dot color/size encodes effect of variable, n = ", nrow(df_tsne))) +
  custom_theme(aspect = 1) +
  scale_color_gradientn(limits = c(-5, 5),
    colours = c(custom_colors[1], grey(0.6, 0.8), custom_colors[2])) +
  scale_size_continuous(range = c(0.5, 7)) +
  geom_text_repel(size = 3, max.overlaps = 50) +
  facet_wrap( ~ treatment, ncol = 2)

print(plot_tsne_linreg)

This strategy reveals a list of interesting condition-specific genes:

• Nitrogen limitation: ssr3532 - unknown short protein, strongest known interaction in STRING with GlsA glutaminase
• Fluctuating light:
  • sll1521 - Putative diflavin flavoprotein A3 (dfa3), negatively corr. with fitness
  • sll0217 Putative diflavin flavoprotein A2 (dfa2), positively corr. with fitness
• Mixotrophy:
  • sll0593 - glk, glucokinase, catalyzes P-ylation of Glc to G6P
  • sll1533 - pilT, fimbria assembly, mobility, Glc transport or sensing?
  • ssl3364 - unknown short protein, strongly interacts with RbcX, RbcR, Prk. Important for C-metabolism adapation?
• Light:
  • ssr2142 ycf19, short unknown protein, interacts with psbO and Tat membrane protein insertion system,
  • slr0963 sir, sulfite reductase, ferredoxin H2O + HS + ferredoxin <-> H+ + reduced ferredoxin + sulfite, strongly interacts with other proteins in sulfur metabolism, specifically related to cofactor biosynthesis, cobalamin (vitamin B12) and siroheme
• Light, mixotrophy, heterotrophy: cluster of photosynthesis related genes increase fitness when KOed: apcA,D,E, psbB,C,D
• Carbon:
  • sll0217 Putative diflavin flavoprotein A2 (dfa2), KO negatively correlated with fitness with C, positive with +FL
  • sll0218 same behavior as dfa2, interacts with dfa2,4, contributes to PSII stabilization, Bersanini et al., 2017

8.4 List of genes with strong fitness correlation

The table with linear regression coefficients and p-values is reshaped to long format for better readability. The kableExtra package is used to color cells for easier recognition. Then we subset the table for each treatment in order to spot the most interesting genes.

df_linreg_wide <- df_linreg %>%
  pivot_wider(names_from = treatment, values_from = coefficient) %>%
  left_join(select(df_gene, locus, sgRNA_target) %>% distinct, by = "locus") %>%
  select(-matches("intercept")) %>%
  filter(if_any(matches("^(carb|light|\\-|\\+)"), ~ abs(.) > 2)) %>%
  mutate(across(matches("carb|light|\\-|\\+"), ~ round(., 3))) %>% 
  ungroup %>% select(sgRNA_target, locus, matches("."))

color_table <- function(df, variable) {
  filter(df, abs(.data[[variable]]) > 2) %>% 
  select(matches("^(sg|loc|r_s|carb|light|\\-|\\+)") | all_of(paste0("pval_", variable))) %>%
  arrange(desc(.data[[variable]])) %>%
  mutate(across(3:8, ~ cell_spec(., "html", color = "white",
      background = spec_color(., option = "E", scale = c(-5.5, 5.5)),
      bold = TRUE))) %>%
  kbl(format = "html", escape = F) %>%
  kable_paper("striped", full_width = F)
}

df_linreg_wide %>% color_table("carbon")

sgRNA_target	locus	carbon	light	-N	+FL	+G	+D	r_squared	pval_carbon
sll0364	sll0364	2.834	-2.812	-0.108	1.758	0.601	3.578	0.8717770	0.024
slr1095	slr1095	2.429	-2.1	-1.284	-1.592	-0.274	1.777	0.6370579	0.083
sll1734	sll1734	2.426	-1.267	-0.478	-1.625	0.837	1.183	0.6560893	0.099
slr0211	slr0211	2.354	0.469	-0.044	-0.362	0.786	1.884	0.8217569	0.022
ftsZ	sll1633	2.34	-2.314	-1.677	-1.439	0.318	2.136	0.9249829	0.006
ndhD3	sll1733	2.326	-1.213	-0.489	-1.392	0.67	1.172	0.6640536	0.085
ssr3532	ssr3532	2.303	-1.661	-3.733	-1.464	0.702	1.64	0.6071408	0.163
ndhF2	sll1732	2.163	-0.874	-0.602	-1.431	1.022	1.124	0.6649520	0.103
slr1818	slr1818	2.159	-1.144	-1.369	-1.55	-0.087	1.675	0.5831706	0.116
sll0488	sll0488	2.151	-1.322	-1.418	-1.434	0.047	1.624	0.6312515	0.090
sll0481	sll0481	2.141	-3.371	-1.601	-1.335	0.485	1.056	0.9686290	0.002
sll0995	sll0995	2.101	-0.278	-1.933	-1.322	0.722	1.384	0.6077491	0.132
cmpB	slr0041	2.041	-1.904	-0.766	-1.032	0.133	1.15	0.5577051	0.126
sir	slr0963	-2.034	4.169	1.41	1.489	-0.345	-0.303	0.8364473	0.048
rpl24	sll1807	-2.081	0.994	0.573	1.814	0.023	-1.225	0.4871941	0.200
rps9	sll1822	-2.087	0.874	0.64	1.319	-0.719	-0.988	0.5755515	0.137
slr6107	slr6107	-2.108	1.215	1.13	1.151	-0.237	-1.246	0.5493324	0.124
slr0272	slr0272	-2.122	1.596	1.27	1.27	-1.038	-0.791	0.6159525	0.127
leuD	sll1444	-2.209	2.505	-0.124	1.64	-1.451	-0.038	0.6714334	0.146
rps17	ssl3437	-2.241	0.79	1.027	1.922	-0.148	-1.382	0.5117852	0.184
slr0007	slr0007	-2.308	1.932	0.695	2.012	-0.851	-1.068	0.6792795	0.109
slr1938	slr1938	-2.311	0.268	0.592	1.075	-0.772	-1.967	0.7489109	0.051
rpl35	ssl1426	-2.361	1.851	1.677	1.293	-0.482	-1.844	0.5899106	0.124
sll0218	sll0218	-2.376	1.766	0.727	1.766	-0.193	-1.1	0.6105884	0.110
sll0217	sll0217	-2.418	1.723	0.954	2.228	-0.286	-1.152	0.6281035	0.118
sll0933	sll0933	-2.634	1.271	0.081	2.115	-0.633	-0.702	0.6558952	0.105
slr1245	slr1245	-2.694	1.826	1.245	-0.805	1.092	-3.224	0.8325233	0.019
rps15	ssl1784	-2.769	1.647	1.312	2.437	-0.758	-1.499	0.7223751	0.075

df_linreg_wide %>% color_table("light")

sgRNA_target	locus	carbon	light	-N	+FL	+G	+D	r_squared	pval_light
apcE	slr0335	-0.681	5.071	0.314	1.093	3.962	1.65	0.8255851	0.055
apcA	slr2067	0.027	4.441	0.576	0.809	3.301	2.922	0.7380667	0.127
sll1878	sll1878	-1.585	4.291	2.016	2.165	1.301	-0.213	0.7988744	0.024
sir	slr0963	-2.034	4.169	1.41	1.489	-0.345	-0.303	0.8364473	0.031
cpcB	sll1577	0.12	4.032	0.141	0.74	2.544	2.476	0.8276584	0.053
murC	slr1423	0.232	3.875	0.345	0.027	1.164	-0.248	0.6419118	0.096
sll1378	sll1378	-0.955	3.701	-0.4	1.2	2.782	0.937	0.9545533	0.006
sll1879	sll1879	-0.306	3.662	-0.894	-0.217	0.681	-0.74	0.7312512	0.074
hitB	slr0327	-1.726	3.606	1.56	1.726	0.576	-0.39	0.6687398	0.087
cpcA	sll1578	0.11	3.57	0.261	0.609	2.255	2.162	0.8434275	0.044
slr0947	slr0947	0.18	3.422	0.454	0.742	1.52	0.613	0.7928550	0.028
slr1990	slr1990	-0.768	3.219	0.197	0.994	2.472	2.429	0.8920252	0.044
slr1102	slr1102	0.056	3.162	-0.118	0.511	2.737	0.819	0.8207709	0.065
amiC	slr0447	-1.235	3.159	0.549	-0.175	1.309	0.402	0.9300166	0.007
sll6055	sll6055	-1.122	3.123	0.558	0.514	2.406	2.128	0.8376624	0.091
cyp2	slr0574	-0.6	3.073	0.867	1.027	2.347	-0.438	0.9444943	0.003
sll1945	sll1945	-0.194	3.072	0.642	0.625	0.063	1.299	0.7068087	0.059
sll0689	sll0689	-0.125	3.062	0.19	0.41	0.027	0.162	0.5002592	0.178
narB	sll1454	-0.223	3.011	0.044	0.014	0.034	0.634	0.8309629	0.025
def	slr1549	0.929	2.973	1.091	-0.039	1.821	0.97	0.7394573	0.103
slr7096	slr7096	-0.452	2.972	-0.83	0.696	-0.16	0.433	0.9130593	0.010
psbJ	smr0008	0.192	2.968	0.293	0.419	3.216	3.222	0.8923081	0.090
slr1505	slr1505	-0.682	2.926	0.456	0.864	2.222	3.068	0.9263147	0.032
slr0483	slr0483	0.315	2.922	0.533	0.633	1.047	0.53	0.9363725	0.003
nirA	slr0898	-1.083	2.919	-0.106	0.515	-0.343	-0.004	0.7628800	0.063
apcB	slr1986	0.627	2.876	0.319	0.432	2.241	2.188	0.7619664	0.129
slr0734	slr0734	0.52	2.819	0.607	0.753	1.99	2.183	0.8211958	0.067
slr1170	slr1170	-0.807	2.766	-0.676	0.308	0.668	-0.598	0.7363329	0.070
ycf38	sll0760	-0.445	2.748	-0.294	0.164	0.823	1.469	0.6739376	0.122
ssl0331	ssl0331	-1.177	2.738	1.195	1.389	0.463	-1.187	0.8598838	0.019
slr1693	slr1693	-1.723	2.719	0.576	0.898	-0.397	-1.131	0.8525814	0.045
psbO	sll0427	0.64	2.707	0.222	0.633	3.339	1.65	0.8838887	0.078
psbD	sll0849	-0.861	2.664	1.002	-0.005	3.231	3.101	0.8557861	0.192
slr1692	slr1692	0.437	2.655	0.33	0.811	1.705	1.623	0.8429081	0.039
hemB	sll1994	-0.063	2.652	-0.914	0.697	0.646	0.836	0.6646820	0.110
slr2042	slr2042	-0.364	2.633	0.809	1.148	1.345	-1.302	0.8793763	0.008
cpcG	slr2051	-0.205	2.621	0.183	0.472	2.294	1.485	0.9132484	0.027
slr1302	slr1302	-0.999	2.591	0.061	0.454	1.365	-0.191	0.8534901	0.026
moeB	sll1536	-1.04	2.579	1.069	0.234	-0.366	-0.368	0.7776446	0.054
sll0148	sll0148	0.357	2.556	-0.37	-0.143	1.982	0.077	0.7719391	0.094
plsX	slr1510	0.976	2.549	0.169	-0.006	0.186	1.868	0.6005493	0.205
sll6109	sll6109	-0.938	2.516	0.586	0.717	0.079	0.159	0.8704727	0.013
cysH	slr1791	-1.968	2.512	0.674	1.242	0.068	-0.998	0.8428947	0.057
leuD	sll1444	-2.209	2.505	-0.124	1.64	-1.451	-0.038	0.6714334	0.313
trxA3	slr0623	-1.336	2.493	1.078	1.881	0.16	-0.637	0.8097690	0.057
nrtB	sll1451	-0.107	2.485	0.272	0.388	-0.244	1.229	0.8884861	0.012
slr1841	slr1841	-0.166	2.451	1.171	0.655	0.131	0.426	0.7466448	0.040
sll2003	sll2003	-0.423	2.396	0.519	0.236	1.625	0.597	0.8779608	0.022
drgA	slr1719	-1.649	2.368	0.596	1.457	-0.317	-0.446	0.6695048	0.187
sll1380	sll1380	-0.367	2.364	0.551	0.741	1.707	0.674	0.9376518	0.006
petH	slr1643	-1.23	2.351	1.048	1.592	2.039	-2.97	0.7531722	0.109
sll1500	sll1500	-1.256	2.346	-0.004	0.248	-0.861	-0.634	0.7141056	0.150
sll1304	sll1304	-0.655	2.334	0.667	0.558	0.589	0.622	0.8743538	0.010
sll0301	sll0301	0.045	2.33	-0.396	-0.094	2.231	0.605	0.9108020	0.034
slr0950	slr0950	-0.06	2.276	0.509	0.314	1.55	0.089	0.8557459	0.020
hemC	slr1887	-0.702	2.265	0.206	0.998	0.118	0.461	0.6255683	0.111
trpF	sll0356	0.03	2.257	0.118	0.505	-0.803	1.243	0.7586823	0.082
ndhF	slr2009	0.221	2.237	0.905	0.613	2.258	3.08	0.9355070	0.043
slr1042	slr1042	0.571	2.208	0.633	0.498	0.785	0.821	0.5506013	0.163
proC	slr0661	-0.746	2.205	0.115	-0.077	1.399	1.48	0.8763769	0.057
rpiA	slr0194	-0.278	2.194	1.674	1.505	2.476	-1.452	0.9605876	0.003
thrA	sll0455	0.548	2.169	-0.427	-0.459	0.803	0.359	0.6473974	0.170
glnA	slr1756	-0.699	2.114	0.035	0.284	-1.379	0.004	0.8295902	0.089
slr0519	slr0519	-1.506	2.09	1.043	1.988	0.287	-1.1	0.6946988	0.190
ssl1918	ssl1918	-0.063	2.086	0.326	0.449	1.257	0.418	0.7483282	0.054
sll0847	sll0847	-0.875	2.075	-0.013	0.095	0.622	0.374	0.7363211	0.083
ccmK4	slr1839	-0.233	2.073	0.584	0.413	2.248	0.326	0.7885394	0.106
pdhB	sll1721	-0.862	2.072	1.323	1.246	0.616	-0.273	0.7513861	0.044
sll1025	sll1025	0.419	2.061	0.377	0.473	1.319	0.495	0.7599508	0.059
prc	slr1751	0.376	2.057	1.334	0.611	0.429	0.649	0.5909157	0.136
slr0771	slr0771	-0.084	2.042	0.311	0.342	0.239	-0.095	0.7574635	0.037
tufA	sll1099	-1.461	2.041	1.399	2.218	0.036	-1.078	0.7057584	0.217
fabF	sll1069	0.361	2.04	1.575	0.069	0.345	-0.025	0.8170498	0.048
aroB	slr2130	-1.206	2.02	-0.656	0.772	-0.523	-0.045	0.8102623	0.094
nrtA	sll1450	-0.214	2.019	0.246	0.238	-0.409	1.571	0.8602072	0.026
hhoB	sll1427	-0.04	2.017	-0.198	0.237	2.207	0.158	0.7804217	0.129
slr0813	slr0813	1.839	-2.009	0.304	0.354	-0.981	0.595	0.7628332	0.187
slr1783	slr1783	-0.973	-2.014	-0.805	-0.173	-0.65	0.265	0.8726587	0.038
sll0176	sll0176	-0.408	-2.085	-2.021	-1.717	-1.526	0.952	0.6551709	0.151
slr1095	slr1095	2.429	-2.1	-1.284	-1.592	-0.274	1.777	0.6370579	0.325
clpP4	sll0534	-0.193	-2.123	-1.24	-0.652	-0.922	-0.194	0.7367912	0.042
sll5063	sll5063	-1.163	-2.128	-0.35	0.218	-0.272	0.018	0.8440277	0.070
mraY	sll0657	1.211	-2.188	-0.758	-0.99	-0.843	0.787	0.8718734	0.017
sll0162	sll0162	0.229	-2.203	-0.403	-0.234	-0.405	0.062	0.6347022	0.078
fabF2	slr1332	0.401	-2.251	0.251	0.187	-2.097	0.38	0.9209766	0.021
slr1098	slr1098	0.903	-2.261	-0.199	-0.434	0.479	-2.543	0.4195152	0.403
sll1757	sll1757	0.783	-2.271	-0.494	-0.752	-1.562	-0.125	0.5383796	0.200
ftsZ	sll1633	2.34	-2.314	-1.677	-1.439	0.318	2.136	0.9249829	0.039
gcvP	slr0293	1.861	-2.335	-1.541	-0.695	-0.558	1.696	0.9451148	0.007
slr0484	slr0484	0.087	-2.342	-0.79	0.038	-0.367	-0.061	0.7395823	0.049
ssr0657	ssr0657	0.935	-2.358	-0.659	-0.992	-2.021	1.496	0.9170055	0.009
sll1915	sll1915	0.381	-2.385	-0.147	-0.431	-1.437	-0.118	0.6397709	0.110
dnaJ4	sll0897	0.955	-2.471	-0.087	0.165	-0.67	1.728	0.9304110	0.007
psaK	ssr0390	1.17	-2.479	-1.404	-0.784	-0.729	0.717	0.6971797	0.062
ppiB	sll0227	0.735	-2.528	-1.035	-0.173	-0.759	0.306	0.8520480	0.013
slr0643	slr0643	0.127	-2.545	-0.449	0.098	-0.647	0.279	0.5529642	0.140
sll0498	sll0498	0.407	-2.577	-1.166	-1.108	-2.148	-1.967	0.8482665	0.063
mrdB	slr1267	0.179	-2.655	-0.252	-0.24	-1.365	0.704	0.8603778	0.014
rpoE	slr1545	0.874	-2.759	-0.425	0.6	-1.246	1.277	0.5615657	0.194
sll0364	sll0364	2.834	-2.812	-0.108	1.758	0.601	3.578	0.8717770	0.119
clpX	sll0535	0.736	-2.966	-1.054	-1.121	-1.503	0.01	0.7464408	0.035
sll0481	sll0481	2.141	-3.371	-1.601	-1.335	0.485	1.056	0.9686290	0.002
sll0877	sll0877	1.76	-3.694	-2	-1.39	-1.273	0.847	0.7839636	0.032
ycf19	ssr2142	1.071	-3.914	-0.278	0.284	-2.735	1.742	0.5878787	0.184

df_linreg_wide %>% color_table("-N")

sgRNA_target	locus	carbon	light	-N	+FL	+G	+D	r_squared	pval_-N
atpA	sll1326	-0.989	-0.069	2.047	0.064	0.272	-1.903	0.3425321	0.396
sll1878	sll1878	-1.585	4.291	2.016	2.165	1.301	-0.213	0.7988744	0.179
sll0176	sll0176	-0.408	-2.085	-2.021	-1.717	-1.526	0.952	0.6551709	0.168
slr1079	slr1079	1.742	-1.553	-2.022	-1.193	-0.394	1.843	0.5565884	0.304
sll5046	sll5046	1.34	-1.273	-2.118	-0.985	-0.329	1.365	0.6360797	0.152
ssr3532	ssr3532	2.303	-1.661	-3.733	-1.464	0.702	1.64	0.6071408	0.201

df_linreg_wide %>% color_table("+FL")

sgRNA_target	locus	carbon	light	-N	+FL	+G	+D	r_squared	pval_+FL
rps15	ssl1784	-2.769	1.647	1.312	2.437	-0.758	-1.499	0.7223751	0.204
pyrG	sll1443	-1.738	0.567	-0.111	2.307	-1.107	-0.398	0.6651024	0.212
sll0217	sll0217	-2.418	1.723	0.954	2.228	-0.286	-1.152	0.6281035	0.258
tufA	sll1099	-1.461	2.041	1.399	2.218	0.036	-1.078	0.7057584	0.112
sll1878	sll1878	-1.585	4.291	2.016	2.165	1.301	-0.213	0.7988744	0.085
sll0933	sll0933	-2.634	1.271	0.081	2.115	-0.633	-0.702	0.6558952	0.294
fus2	sll1098	-1.02	1.482	0.038	2.077	0.13	-0.815	0.7239763	0.087
slr0007	slr0007	-2.308	1.932	0.695	2.012	-0.851	-1.068	0.6792795	0.267
ribC	sll0300	-1.073	-0.365	-0.267	-2.057	-0.721	-0.724	0.7973653	0.037
entC	slr0817	-0.511	-0.844	-0.456	-2.123	2.558	0.74	0.9773035	0.009
ribA	sll1894	-0.927	-0.39	-0.139	-2.37	-0.725	-0.64	0.6200967	0.101
sll1521	sll1521	-0.353	-0.754	-0.52	-3.993	-0.671	-0.342	0.9636366	0.001

df_linreg_wide %>% color_table("+G")

sgRNA_target	locus	carbon	light	-N	+FL	+G	+D	r_squared	pval_+G
apcE	slr0335	-0.681	5.071	0.314	1.093	3.962	1.65	0.8255851	0.056
dgt	sll0398	-0.463	0.606	0.269	1.052	3.782	-0.261	0.9841219	0.000
psbO	sll0427	0.64	2.707	0.222	0.633	3.339	1.65	0.8838887	0.021
apcA	slr2067	0.027	4.441	0.576	0.809	3.301	2.922	0.7380667	0.142
psbC	sll0851	-0.001	1.691	0.24	0.281	3.288	3.75	0.9364716	0.019
sll1496	sll1496	1.499	-1.248	-0.507	-0.657	3.237	-2.22	0.9150619	0.013
psbD	sll0849	-0.861	2.664	1.002	-0.005	3.231	3.101	0.8557861	0.072
psbJ	smr0008	0.192	2.968	0.293	0.419	3.216	3.222	0.8923081	0.037
slr0758	slr0758	-0.644	1.944	0.251	-0.176	2.791	-1.499	0.9317621	0.005
sll1378	sll1378	-0.955	3.701	-0.4	1.2	2.782	0.937	0.9545533	0.006
slr1102	slr1102	0.056	3.162	-0.118	0.511	2.737	0.819	0.8207709	0.050
sll0556	sll0556	1.268	0.982	0.25	-0.377	2.727	0.121	0.7859377	0.060
slr2070	slr2070	0.846	-0.597	-0.408	-0.608	2.711	-1.752	0.7416662	0.075
cbbA	sll0018	-0.644	1.164	0.415	0.545	2.607	2.054	0.9537621	0.008
entC	slr0817	-0.511	-0.844	-0.456	-2.123	2.558	0.74	0.9773035	0.005
cpcB	sll1577	0.12	4.032	0.141	0.74	2.544	2.476	0.8276584	0.093
ssr2062	ssr2062	0.5	0.071	0.167	-0.086	2.528	-1.616	0.9322018	0.004
rpiA	slr0194	-0.278	2.194	1.674	1.505	2.476	-1.452	0.9605876	0.001
slr1990	slr1990	-0.768	3.219	0.197	0.994	2.472	2.429	0.8920252	0.047
trx	sll1057	-0.491	1.161	-0.106	0.132	2.417	0.6	0.9611652	0.003
sll6055	sll6055	-1.122	3.123	0.558	0.514	2.406	2.128	0.8376624	0.094
psbH	ssl2598	0.265	1.696	0.215	-0.134	2.349	1.235	0.8920776	0.024
cyp2	slr0574	-0.6	3.073	0.867	1.027	2.347	-0.438	0.9444943	0.003
ccsA	sll1513	0.287	1.628	0.096	0.428	2.321	0.814	0.9010260	0.013
cpcC2	sll1579	-0.036	1.144	0.101	0.224	2.32	0.641	0.9783295	0.001
cpcG	slr2051	-0.205	2.621	0.183	0.472	2.294	1.485	0.9132484	0.019
sll0062	sll0062	0.605	1.181	0.165	0.639	2.262	1.004	0.8522771	0.027
ndhF	slr2009	0.221	2.237	0.905	0.613	2.258	3.08	0.9355070	0.020
cpcA	sll1578	0.11	3.57	0.261	0.609	2.255	2.162	0.8434275	0.079
ccmK4	slr1839	-0.233	2.073	0.584	0.413	2.248	0.326	0.7885394	0.045
apcB	slr1986	0.627	2.876	0.319	0.432	2.241	2.188	0.7619664	0.130
sll0301	sll0301	0.045	2.33	-0.396	-0.094	2.231	0.605	0.9108020	0.018
slr1505	slr1505	-0.682	2.926	0.456	0.864	2.222	3.068	0.9263147	0.035
hhoB	sll1427	-0.04	2.017	-0.198	0.237	2.207	0.158	0.7804217	0.056
rub	slr2033	0.214	1.429	0.237	-0.416	2.109	1.991	0.9937546	0.000
petH	slr1643	-1.23	2.351	1.048	1.592	2.039	-2.97	0.7531722	0.085
ssr0657	ssr0657	0.935	-2.358	-0.659	-0.992	-2.021	1.496	0.9170055	0.007
fabF2	slr1332	0.401	-2.251	0.251	0.187	-2.097	0.38	0.9209766	0.012
sll0498	sll0498	0.407	-2.577	-1.166	-1.108	-2.148	-1.967	0.8482665	0.053
slr1974	slr1974	-1.295	-0.202	0.509	0.64	-2.153	-0.861	0.7460569	0.127
glcP	sll0771	0.113	-0.381	-0.143	0.008	-2.164	-1.027	0.9982143	0.000
minE	ssl0546	0.975	-1.768	-1.032	-0.849	-2.194	0.18	0.8244200	0.032
purA	sll1823	-1.4	-0.755	1.149	0.416	-2.4	-0.727	0.6987707	0.151
glgA	sll1393	-0.137	-0.226	0.031	-0.204	-2.7	1.505	0.9275690	0.004
pilT	sll1533	-1.207	-0.456	-0.979	1.343	-2.71	0.646	0.7070054	0.138
ycf19	ssr2142	1.071	-3.914	-0.278	0.284	-2.735	1.742	0.5878787	0.226
ssl3364	ssl3364	0.214	-1.289	-0.623	-0.271	-3.033	-0.61	0.9116291	0.010
glk	sll0593	-0.008	-0.723	-0.614	-0.394	-3.701	-2.08	0.9860610	0.000

df_linreg_wide %>% color_table("+D")

sgRNA_target	locus	carbon	light	-N	+FL	+G	+D	r_squared	pval_+D
psbB	slr0906	-0.064	1.707	0.622	-0.176	1.44	4.62	0.8709903	0.028
psbE	ssr3451	0.191	1.559	0.07	0.163	1.443	3.846	0.9215596	0.012
psbC	sll0851	-0.001	1.691	0.24	0.281	3.288	3.75	0.9364716	0.030
psbF	smr0006	0.201	1.328	0.045	-0.021	1.078	3.584	0.9289141	0.009
sll0364	sll0364	2.834	-2.812	-0.108	1.758	0.601	3.578	0.8717770	0.072
rpoDI	slr0653	1.855	-0.679	-1.12	-1.39	-1.275	3.412	0.5935773	0.155
psbD2	slr0927	-0.005	1.286	0.228	-0.241	1.923	3.353	0.9093335	0.031
psbJ	smr0008	0.192	2.968	0.293	0.419	3.216	3.222	0.8923081	0.079
psbD	sll0849	-0.861	2.664	1.002	-0.005	3.231	3.101	0.8557861	0.152
ndhF	slr2009	0.221	2.237	0.905	0.613	2.258	3.08	0.9355070	0.018
slr1505	slr1505	-0.682	2.926	0.456	0.864	2.222	3.068	0.9263147	0.031
tktA	sll1070	0.198	0.59	0.503	0.822	1.981	2.992	0.9639206	0.006
apcA	slr2067	0.027	4.441	0.576	0.809	3.301	2.922	0.7380667	0.289
lysA	sll0504	-0.206	0.251	0.394	0.088	0.396	2.767	0.9715674	0.001
cpcB	sll1577	0.12	4.032	0.141	0.74	2.544	2.476	0.8276584	0.181
slr1990	slr1990	-0.768	3.219	0.197	0.994	2.472	2.429	0.8920252	0.102
psbL	smr0007	0.139	1.635	0.417	0.57	1.385	2.402	0.8022242	0.083
apcB	slr1986	0.627	2.876	0.319	0.432	2.241	2.188	0.7619664	0.232
slr0734	slr0734	0.52	2.819	0.607	0.753	1.99	2.183	0.8211958	0.134
hemA	slr1808	0.905	1.077	-0.68	-0.639	1.149	2.174	0.5761540	0.327
cpcA	sll1578	0.11	3.57	0.261	0.609	2.255	2.162	0.8434275	0.164
ftsZ	sll1633	2.34	-2.314	-1.677	-1.439	0.318	2.136	0.9249829	0.054
sll6055	sll6055	-1.122	3.123	0.558	0.514	2.406	2.128	0.8376624	0.217
sll0930	sll0930	0.084	0.116	0.329	0.368	1.209	2.055	0.8790537	0.046
cbbA	sll0018	-0.644	1.164	0.415	0.545	2.607	2.054	0.9537621	0.043
sll0983	sll0983	1.392	-1.936	0.075	-0.306	-0.69	2.036	0.8763437	0.041
glk	sll0593	-0.008	-0.723	-0.614	-0.394	-3.701	-2.08	0.9860610	0.012
ndhC	slr1279	0.96	-0.173	0.393	-0.18	-0.985	-2.147	0.8635036	0.098
ycf43	sll0194	-0.722	-0.119	0.025	0.102	1.626	-2.206	0.9735251	0.001
sll1496	sll1496	1.499	-1.248	-0.507	-0.657	3.237	-2.22	0.9150619	0.088
pmgA	sll1968	1.162	-1.905	0.387	-1.343	-1.769	-2.242	0.5849242	0.429
ssr2333	ssr2333	0.305	-1.552	-0.003	-0.034	0.305	-2.319	0.4255312	0.285
ssl0438	ssl0438	0.12	-1.474	-0.18	-0.419	-0.219	-2.375	0.8654067	0.019
atpI	sll1322	-1.547	-0.422	1.294	0.662	0.252	-2.475	0.4121488	0.353
slr1098	slr1098	0.903	-2.261	-0.199	-0.434	0.479	-2.543	0.4195152	0.367
talB	slr1793	0.152	-0.144	-0.806	-0.023	0.111	-2.879	0.9813772	0.000
petH	slr1643	-1.23	2.351	1.048	1.592	2.039	-2.97	0.7531722	0.066
zwf	slr1843	-0.079	0.297	-0.149	0.054	-0.043	-2.998	0.9966699	0.000
slr1245	slr1245	-2.694	1.826	1.245	-0.805	1.092	-3.224	0.8325233	0.067

Based on the multiple linear model correlations, we can try to extract a shortlist of the most interesting hypothetical genes. These could warrant further investigations.

list_top_unknown_hits <- df_linreg_wide %>%
  left_join(df_uniprot, by = "locus") %>%
  # filter by name: only unknown proteins
  filter(
    is.na(gene_name_short),
    str_detect(protein, "[a-zA-Z]{3}[0-9]{4} protein|Uncharacterized")) %>%
  # filter by effect: only correlation > 3
  filter(if_any(matches("^(carb|light|\\-|\\+)"), ~ abs(.) > 3)) %>%
  arrange(desc(r_squared)) %>%
  pull(locus)

df_linreg_wide %>% filter(locus %in% list_top_unknown_hits) %>%
  select(!starts_with("pval"), -sgRNA_target) %>%
  mutate(across(2:7, ~ cell_spec(., "html", color = "white",
      background = spec_color(., option = "E", scale = c(-5.5, 5.5)),
      bold = TRUE))) %>%
  kbl(format = "html", escape = F) %>%
  kable_paper("striped", full_width = F)

locus	carbon	light	-N	+FL	+G	+D	r_squared
sll0364	2.834	-2.812	-0.108	1.758	0.601	3.578	0.8717770
sll0481	2.141	-3.371	-1.601	-1.335	0.485	1.056	0.9686290
sll0877	1.76	-3.694	-2	-1.39	-1.273	0.847	0.7839636
sll1378	-0.955	3.701	-0.4	1.2	2.782	0.937	0.9545533
sll6055	-1.122	3.123	0.558	0.514	2.406	2.128	0.8376624
slr1102	0.056	3.162	-0.118	0.511	2.737	0.819	0.8207709
slr1505	-0.682	2.926	0.456	0.864	2.222	3.068	0.9263147
slr1990	-0.768	3.219	0.197	0.994	2.472	2.429	0.8920252
ssl3364	0.214	-1.289	-0.623	-0.271	-3.033	-0.61	0.9116291
ssr3532	2.303	-1.661	-3.733	-1.464	0.702	1.64	0.6071408

8.5 Extract and analyze interesting gene clusters

The list above shows the genes whose fitness is most significantly correlated with one of the treatments. This list of genes is extracted and then simply fitness per condition is plotted as a heatmap, in order to confirm the trends from fitting the multiple liner regression models.

library(ggheatmap)

plot_sgRNAs_light <- df_gene %>%
  
  # make df with list of strongest correlated genes
  filter(locus %in% list_top_unknown_hits) %>%
  
  # reshape to wide format
  mutate(wmean_fitness = wmean_fitness %>% replace(., . > 3.5, 3.5) %>% replace(., . < -3.5, -3.5)) %>%
  filter(time == 0) %>% select(sgRNA_target, condition, wmean_fitness) %>%
  pivot_wider(names_from = condition, values_from = wmean_fitness) %>%
  column_to_rownames("sgRNA_target") %>% t %>%
  
  # plot heatmap
  ggheatmap(cluster_rows = TRUE, cluster_cols = TRUE,
    color = brewer_pal(direction = -1, palette = "RdBu")(5),
    show_cluster_rows   = FALSE, cluster_num = c(1,1),
    tree_color_cols = grey(0.5),
  ) %>% ggheatmap_theme(plotlist = 1, theme = list(
    custom_theme() + theme(axis.text.x = element_text(angle = 90, vjust = 0.5), 
      axis.title.x = element_blank(), axis.title.y = element_blank())
  ))

Warning: `guides(<scale> = FALSE)` is deprecated. Please use `guides(<scale> = "none")` instead.

print(plot_sgRNAs_light)

save_plot(plot_sgRNAs_light, width = 8.0, height = 3.5)

Summary

• sll0364 - 139 AA. KD has higher fitness in HC conditions and lower fitness in HL. Negatively regulating carbon metabolism?
• sll0481 - 155 AA. KD has higher fitness in +G conditions and lower fitness in HL. Membrane localization. Negatively regulating glycolysis?
• sll0877 - 456 AA. KD has higher fitness only in HC,LL. Mitigates light limitation?
• ssl3364 - 74 AA. KD has lower fitness on all HC/+G conditions. This protein is known as CP12 protein, regulating glycolytic flux at GAPDH and PRK.
• ssr3532 - 80 AA. KD lower fitness on N-limitation and C-limitation (LC-HL combinations). Same operon as glutaminase glsA (slr2079, catalyzes deamination of gln –> glu), regulatory, involved in N metabolism?
• slr1990 - 240 AA, 5 TM domains. KD higher fitness in photoheterotrophy, lower fitness in all HC/LL conditions. Something important for photosystems? Something that wastes e- in photoheterotrophic conditions?
• sll6055 - 152 AA. Fitness profile as above. Multiubiquitin domain, involved in protein modification/degradation of PS proteins?
• slr1505 - 198 AA. Fitness profile as above. No useful information.
• sll1378 - 300 AA. KD has lower fitness on all LL conditions. Membrane associated protein? In STRING, potential interaction with PbsA1 and PbsA2 (Heme oxygenase 1 and 2). Potentially important for chlorophyll or heme biosynthesis –> would explain importance for photosynthesis in LL condition.
• slr1102 - 853 AA. KD has lower fitness on all LL conditions. 4 known domains, FHA (forkhead-associated domain is a phosphopeptide recognition domain found in many regulatory proteins), PAS (signaling, often involved in circadian proteins, detect their signal by way of an associated cofactor like heme, flavin), GGDEF (involved in signal transduction, likely to catalyze synthesis or hydrolysis of cyclic diguanylate c-diGMP), EAL (shown to stimulate degradation of a second messenger, cyclic di-GMP, candidate for a diguanylate phosphodiesterase function. Together with the GGDEF domain, EAL might be involved in regulating cell surface adhesiveness in bacteria). Source: InterPro. Embedded in a tight network of interacting proteins all involved in chromophore biosynthesis/maturation.

Apc and cpc repression mutants encoding phycobilisomes are also enriched in high light

plot_sgRNAs_phycobil <- df_gene %>%
  filter(str_detect(gene_name, "[ac]pc"), time == 0) %>%
  mutate(wmean_fitness = wmean_fitness %>% replace(., . > 4, 4) %>% replace(., . < -4, -4)) %>%
  ggplot(aes(x = sgRNA_target, y = condition, fill = wmean_fitness)) +
  geom_tile() + custom_theme() +
  labs(title = "Apc/Cpc mutants enriched in high light/CO2", x = "", y = "") +
  theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1)) +
  scale_fill_distiller(palette = "RdBu", limits = c(-4, 4))

print(plot_sgRNAs_phycobil)

save_plot(plot_sgRNAs_phycobil, width = 6.5, height = 3.5)

9 Direct comparison of gene fitness

9.1 Fitness of all conditions vs each other

We can plot selected conditions against each other and add gene labels in order to find or confirm particular patterns.

make_fitness_plot <- function(data, vars, title = NULL) {
  # prepare data for two  variables each
  data %>% ungroup %>%
    filter(condition %in% vars, sgRNA_type == "gene") %>%
    select(locus, sgRNA_target, condition, wmean_fitness) %>% distinct %>%
    pivot_wider(names_from = condition, values_from = wmean_fitness) %>%
    mutate(
      dfit = get(vars[1]) - get(vars[2]),
      significant = !between(dfit, quantile(dfit, probs = c(0.003)),
        quantile(dfit, probs = c(0.997))),
      sgRNA_target = if_else(significant, sgRNA_target, "")) %>%
    
    # plot
    ggplot(aes(x = get(vars[1]), y = get(vars[2]), 
      color = significant, label = sgRNA_target)) +
    geom_point(size = 1) + custom_theme(legend.position = 0) +
    geom_abline(intercept = 0, slope = 1, col = grey(0.5), lty = 2, size = 0.8) +
    geom_abline(intercept = 4, slope = 1, col = grey(0.5), lty = 2, size = 0.8) +
    geom_abline(intercept = -4, slope = 1, col = grey(0.5), lty = 2, size = 0.8) +
    geom_text_repel(size = 3, max.overlaps = 50) +
    labs(title = title, x = vars[1], y = vars[2]) +
    coord_cartesian(xlim = c(-9, 5), ylim = c(-9, 5)) +
    scale_color_manual(values = c(grey(0.5), custom_colors[2]))
}

# browse through all possible condition combinations;
# we need a helper function that detects duplicated combinations
duplicated_2vec <- function(x, y) {
  xy = paste(x, y); yx = paste(y, x)
  sapply(xy, function(xval) {
    which(xval == yx) <= which(xval == xy)
  })
}

list_condition_pairs <- lapply(
  unique(df_gene$condition) %>% expand_grid(x = ., y = .) %>%
    filter(!duplicated_2vec(x, y)) %>% t %>% as.data.frame %>% as.list,
  function(var) {
    make_fitness_plot(df_gene, vars = var,
      title = paste(var, collapse = "  -  "))
  }
)

# export images
invisible(capture.output(
  lapply(list_condition_pairs, function(pl) {
    pl_name <- paste0("../figures/pairwise_comparisons/plot_", pl$labels$x, "_", pl$labels$y, ".png")
    png(filename = pl_name, width = 800, height = 800, res = 120)
    print(pl)
    dev.off()
  })
))

# example of first 4 combinations
list_condition_pairs[1:4]

$V1

$V2

$V3

$V4

10 Differential fitness of selected gene sets

10.1 Central carbon metabolism

To plot gene fitness for the enzymes of central carbon metabolism, we use the complete list of enzymes and the genes that they are mapped to (obtained from KEGG). We can extract gene sets for specific pathways and plot fitness. We start with glycolysis and Calvin cycle enzymes.

list_central_met_pathways <- c(
  "Glycolysis / Gluconeogenesis",
  "Pentose phosphate pathway",
  "Carbon fixation in photosynthetic organisms",
  "Photosynthesis",
  "Photosynthesis - antenna proteins",
  "Citrate cycle (TCA cycle)",
  "Pyruvate metabolism",
  "Glyoxylate and dicarboxylate metabolism"
)

plot_gene_fitness <- function(df, pw = NULL, gene = NULL, title = NULL) {
  df <- df %>% filter(time == 0)
  if (!is.null(pw)) {
    df <- df %>% inner_join(df_kegg %>% filter(kegg_pathway == pw) %>% select(locus),
      by = "locus")
    title <- pw
  } else if (!is.null(gene)) {
    df <- df %>% filter(locus %in% gene)
  }
  
  ggplot(df, aes(x = condition, y = wmean_fitness, 
    ymin = wmean_fitness-sd_fitness, 
    ymax = wmean_fitness+sd_fitness, fill = condition, color = condition)) +
    geom_col(position = "dodge", width = 0.6) +
    geom_errorbar(position = "dodge", width = 0.6, size = 1) +
    custom_theme(aspect.ratio = 1,
      legend.position = "bottom", legend.key.size = unit(0.4, "cm")) + 
    labs(title = title) +
    theme(axis.text.x = element_blank()) +
    scale_fill_manual(values = colorRampPalette(custom_colors[1:5])(11)) +
    scale_color_manual(values = colorRampPalette(custom_colors[1:5])(11)) +
    facet_wrap(~ sgRNA_target, ncol = 8)
}

print(plot_gene_fitness(df_gene, pw = list_central_met_pathways[[1]]))

ggsave("../figures/plot_fitness_glycolysis.svg",
  plot_gene_fitness(df_gene, pw = list_central_met_pathways[[1]]),
  width = 8, height = 6)

print(plot_gene_fitness(df_gene, pw = list_central_met_pathways[[2]]))

ggsave("../figures/plot_fitness_pentose.svg",
  plot_gene_fitness(df_gene, pw = list_central_met_pathways[[2]]),
  width = 8, height = 5)

print(plot_gene_fitness(df_gene, pw = list_central_met_pathways[[3]]))

ggsave("../figures/plot_fitness_carbonfix.svg",
  plot_gene_fitness(df_gene, pw = list_central_met_pathways[[3]]),
  width = 8, height = 5)

print(plot_gene_fitness(df_gene, pw = list_central_met_pathways[[4]]))

ggsave("../figures/plot_fitness_photosys.svg",
  plot_gene_fitness(df_gene, pw = list_central_met_pathways[[4]]),
  width = 8, height = 11)

print(plot_gene_fitness(df_gene, pw = list_central_met_pathways[[5]]))

ggsave("../figures/plot_fitness_antenna.svg",
  plot_gene_fitness(df_gene, pw = list_central_met_pathways[[5]]),
  width = 8, height = 4)

print(plot_gene_fitness(df_gene, pw = list_central_met_pathways[[6]]))

ggsave("../figures/plot_fitness_citrate.svg",
  plot_gene_fitness(df_gene, pw = list_central_met_pathways[[6]]),
  width = 8, height = 4)

Custom set of stress related genes

OCP (slr1963), pgr5 (ssr2016), Flv1 (sll1521), Flv2 (sll0219), Flv3 (sll0550), Flv4 (sll0217), sll0218 (in flv2/4 operon), SigB (sll0306), SigC (sll0184), SigD (sll2012), SigE (sll1689).

list_stress_genes <- c("slr1963", "ssr2016", "sll1521",
  "sll0219", "sll0550", "sll0217", "sll0218", "sll0306", "sll0184",
  "sll2012", "sll1689")

plot_gene_fitness(df_gene, gene = list_stress_genes, title = "stress related genes")

ggsave("../figures/plot_fitness_stress.svg",
  plot_gene_fitness(df_gene, gene = list_stress_genes, title = "stress related genes"),
  width = 8, height = 4)

Fitness plot for carbon transports and carboxysome shell

slr1512 sbtA, slr1513 sbtB, sll1734 cupA, slr1302 cupB, sll0359 cyabrB1, sll0822 cyabrB2, sll1594 ccmR, sll1031 ccmM, sll1028 ccmK2, sll1029 ccmK1, sll1032 ccmN, slr0436 ccmO, sll1030 ccmL.

list_carboxysome_genes <- c(
  "slr1512", "slr1513", "sll1734", "slr1302", "sll0359", "sll0822", "sll1594",
  "sll1031", "sll1028", "sll1029", "sll1032", "slr0436", "sll1030"
)

plot_gene_fitness(df_gene, gene = list_carboxysome_genes,
  title = "Carboxysome and carbon transporters")

ggsave("../figures/plot_fitness_carboxysome.svg",
  plot_gene_fitness(df_gene, gene = list_carboxysome_genes,
    title = "Carboxysome and carbon transporters"),
  width = 8, height = 4)

Genes whose KD leads to increased fitness

list_genes_pos_fitness <- df_gene %>%
  filter(time == 0, !is.na(locus), wmean_fitness > 2) %>%
  pull(locus) %>% unique

plot_gene_fitness(df_gene, gene = list_genes_pos_fitness, title = "Genes with increased fitness (f > 2)")

ggsave("../figures/plot_fitness_increased.svg",
  plot_gene_fitness(df_gene, gene = list_genes_pos_fitness, title = "Genes with increased fitness (f > 2)"),
  width = 8, height = 8)

Summary: - pmgA is once again the gene with strongest and most widespread fitness increase, validating results from library V1 - slr1916 same phenotype as pmgA just weaker. We also know this one from before. Must have identical role as pmgA. - all PSII genes show increased fitness in photoheterotrophic condition –> PS is a burden here - sll0689, pxcA, slr1609 - all increased fitness in HC,HL, first two are Na+/CO2 (?) trnasporters, slr1609 we know from before, annotated as fatty acid CoA ligase, but probably it’s something different - sll6055, slr1505, slr1990 - all increased fitness in photoheterotrophic condition, and decreased fitness in HC/LL conditions. Not much is known about these genes, probably a role in photosynthesis, as the pattern is similar to psb genes (PSII maturation?) - slr0813, slr0907, slr909, slr1299 - all increased fitness in HC/LL. Not clear what connects these genes functionally.

11 Export summary table of all genes and conditions

Export a summary table of all genes and conditions, so that it’s easy for other people to look up single conditions as for example done in one-by-one fitness comparisons. This is best done in wide format (one column per condition).

df_gene %>% ungroup %>%
  filter(sgRNA_type == "gene") %>%
  select(locus, sgRNA_target, gene_name, condition, wmean_fitness) %>% 
  distinct %>%
  pivot_wider(names_from = condition, values_from = wmean_fitness) %>%
  write_csv("../data/output/fitness_summary.csv")

df_gene %>%
  filter(sgRNA_type == "gene") %>%
  write_csv("../data/output/fitness_genes.csv")

df_kegg %>% write_csv("../data/output/kegg_annotation.csv")

12 Session Info

sessionInfo()